Flights have been grounded, banking systems halted and a whole host of other cyber problems have affected almost everybody in the US this week.
But while there have been some malicious cyber attacks, most of what is happening is far more boring — companies scrambling to prevent bad things happening.
The increased capabilities of artificial intelligence enable it to find vulnerabilities, or “holes” in the world’s software faster than companies can patch them up, and the scramble to fix them is knocking systems offline, a well-placed source told the Post.
For example, last year Microsoft’s security update for November fixed 63 vulnerabilities, according to cybersecurity firm Qualys. This year the same update fixed 1,235.
More than six months ago, Anthropic and rival OpenAI warned that their newest models could find and exploit unknown flaws in major companies’ systems, without being given guidance to. Large companies were then given controlled access to check their systems for bugs and fix them.
The models are so good they turn up dozens of bugs at a time. Finding them was one thing, updating things is another. And most companies’ systems — built on various different software developed at different times over the last three decades and often bolted together — can’t process updates to fix them all at once. That has led to crashes, outages and problems, according to the source.
“Frontier models have crossed the Rubicon on cyber capabilities and are finding copious flaws, which is good for security. But the sheer volume of vulnerabilities is causing a challenge for companies to stay up to date,” the source explained.
“However, we have to remember, ultimately, the patching is making systems less vulnerable.”
Because the models are being handed out on a rolling basis, the source expects more outages over the next few months as more companies finish putting them to work. Updates will also have to come more often in order to keep up.
This week things came to a head before the Thanksgiving break.
Microsoft’s Outlook went dark for nearly 24 hours starting Nov. 12. The next day, the FAA grounded every departing flight in the country for more than four hours. On Monday, Chase bank customers couldn’t move money.
On Tuesday, the National Security Agency’s classified internal network went down.
Fixing a newly found flaw in PHP, the programming language which underpins most websites, “caused some of these systems to not come back up when you patch them as expected,” Director of National Intelligence Jay Clayton told Fox Business.
The flaw surfaced through an OpenAI program that uses AI to hunt for bugs.
Microsoft, the FAA and Chase blamed their own glitches and have made clear there was no cyber attack against them.
Anthropic CEO Dario Amodei has said the pain will pay off. In May, he gave defenders six to 12 months to fix everything before the threat of Chinese developed AI systems could catch up or exploit.
“If we handle this right … we could be in a better position than we started in, because we fixed all these bugs,” he said.
However, not everyone is buying the “just a glitch” explanation.
“Could each individual one be explained by a technical glitch? Yes, to some extent. [But] you never hear of this many things happening in such a tight timeframe,” cybersecurity expert Scott Schober, CEO of Berkeley Varitronics Systems and author of “Senior Cyber,” told the Post.
Clayton said there was no evidence of a coordinated attack or state-sponsored activity, but Schober would only allow it was “possible, not proven yet.”
Thanksgiving travel is already straining aging travel booking systems — providing a real test.
“Transportation systems, the airlines, the backend for booking, radar systems, most of them, unfortunately, are somewhat dated. They are legacy,” Schober said.
“Cyber criminals know this. They are going to take advantage of that. Anytime there’s a disruption or even a glitch, that’s when they’re gonna move in.”
